Process-based karst modelling to relate hydrodynamic and hydrochemical characteristics to system properties

Abstract

More than 30% of Europe's land surface is made up of karst exposures. In some countries, water from karst aquifers constitutes almost half of the drinking water supply. Hydrological simulation models can predict the large-scale impact of future environmental change on hydrological variables. However, the information needed to obtain model parameters is not available everywhere and regionalisation methods have to be applied. The responsive behaviour of hydrological systems can be quantified by individual metrics, so-called system signatures. This study explores their value for distinguishing the dominant processes and properties of five different karst systems in Europe and the Middle East with the overall aim of regionalising system signatures and model parameters to ungauged karst areas. By defining ten system signatures derived from hydrodynamic and hydrochemical observations, a process-based karst model is applied to the five karst systems. In a stepwise model evaluation strategy, optimum parameters and their sensitivity are identified using automatic calibration and global variance-based sensitivity analysis. System signatures and sensitive parameters serve as proxies for dominant processes and optimised parameters are used to determine system properties. To test the transferability of the signatures, they are compared with the optimised model parameters and simple climatic and topographic descriptors of the five karst systems. By sensitivity analysis, the set of system signatures was able to distinguish the karst systems from one another by providing separate information about dominant soil, epikarst, and fast and slow groundwater flow processes. Comparing sensitive parameters to the system signatures revealed that annual discharge can serve as a proxy for the recharge area, that the slopes of the high flow parts of the flow duration curves correlate with the fast flow storage constant, and that the dampening of the isotopic signal of the rain as well as the medium flow parts of the flow duration curves have a non-linear relation to the distribution of groundwater dynamics. Even though, only weak correlations between system signatures and climatic and topographic factors could be found, our approach enabled us to identify dominant processes of the different systems and to provide directions for future large-scale simulation of karst areas to predict the impact of future change on karst water resources

Exploring Controls on Rainfall‐Runoff Events: 2. Regional Patterns and Spatial Controls of Event Characteristics in Germany

“…These catchments are characterized by small subsurface storage (i.e., shallow soils). Possible karstification (Figure e, lower panel) might be an additional source of threshold processes in this cluster (Hartmann et al, ). Small but very frequent rainfall events together with a relatively low dryness index indicate that runoff generation is controlled by the sequence of rainfall events, which modify the extent of the variably saturated region (Berghuijs et al, ).…”

Karst water resources in a changing world: Review of hydrological modeling approaches

“…The tremendous heterogeneity of karst system characteristics means that detailed local observations have to be made to define a catchment or to tune a local model. A range of experimental karst catchments with a more or less intensive monitoring databases are available [ Hartmann et al ., ; White , ], but for most karst systems, the available data are insufficient for the application of mechanistic distributed karst models (see section 4.3). Methods to transfer information from sites with better data availability, like those reviewed for the Predictions in Ungauged Basins initiative [ Blöschl et al ., ; Sivapalan , ], are still insufficient given the specific local evolution each karst system undergoes [ Hartmann et al ., ].…”

“…Hartmann et al . [] found correlations between hydrologic and hydrochemical system signatures and the parameters of a process‐based karst model. However, they did not succeed in regionalizing the signatures with the available predictors (mean temperature, altitude difference, and mean annual precipitation).…”

Bayesian spectral likelihood for hydrological parameter inference

“…[] and Hartmann et al . [] used the streamflow autocorrelation function as a signature to assess the model performance in a multicriteria model identification setting. Schaefli and Zehe [] proposed to assess hydrologic model performance in terms of the Kolmogorov‐Smirnoff distance between the estimated wavelet power spectra of observed and simulated streamflow series.…”